DB-Tracker: Multi-Object Tracking for Drone Aerial Video Based on Box-MeMBer and MB-OSNet

Round 1

Reviewer 1 Report

This manuscript presents the idea of detection & tracking by appearance and position. The results show improvement in detection, tracking, associate & re-id too. The reader finds the idea is innovative with sounds background, theory and results, hence, recommends to accept this manuscript for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

- Better to make the best metrics in bold or colors in Table 1 and 2, so we can find them more easily.

- According to the papers of some existing methods like UAVMOT, Strong SORT, etc, their results are better than the those shown in Table 1 and 2, especially UAVMOT which was also tested on the same dataset VisDrone and UAVDT. Any reasons?

- Follow the previus question, this paper misses some detailed comparison with existing compared methods, like when and why these methods don't work well and why the proposed method is better than them.  

- Miss some qualitative/visual result of the comparison between the proposed DB-Tracker and other existing methods, especially some cases where other methods are worse than the proposed methods (together with analysis of the reason). Better to be on both public dataset and self-collected dataset.

- Would be better and necessary for authors to provide some supplemental video results to prove the continuous consistency and efficiency of the proposed method.

Some grammar, syntax or typo errors:

- I suggest authors to improve the abstract, since it misses the obvious contribution and advantage of the proposed method, but with only technical steps of the pipeline.

- Ref [39] and [40] are exactly the same. 

- Line 521 [3434]

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, "DB-Tracker: Multi-object Tracking for Drone Aerial Video Based on Box-MeMBer and MB-OSNet", the position information of the object is collected by using labelled generalized multi-Bernoulli filter. MB-OSNet is used to extract the appearance features of the objects, and a random finite set method based on the multi-Bernoulli filter is used for multi-object tracking with a nonlinear motion model and a nonlinear observation model.

Technically, the article is good. But it needs to be revised as follows. My recommendation is to accept this revised paper.

1. In the description of the current state of research, the existing research status should be described in more detail so that the reader can understand the advantages and disadvantages of other algorithms.

2. Please write the name of the tool/environment you used for evaluation and the number of percentages, and the methods you used for comparison in the abstract, introduction, and conclusion sections. 

3. In 4.2.1 the YOLOV8 is used to add the attention mechanism to recognize small targets, the lack of experimental comparisons with YOLOV8 can be added with the relevant result graphs during the comparative training process, such as the mAP values of IOUs at 0.5, 0.5:0.95, the Recall curve graphs, and the Precision curve graphs.

4. The conclusion is descriptive and it does not quantitatively the results obtained by numerical and experimental results.

5. Several of the images（Fig.3、4、5） are very low resolution and must be improved.

The Quality of English Language is appropriate.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

- The manuscript's language should be written in passive voice as it is common in the scientific literature. For example, “it is maintained” or “finite sets are mixed” rather than “we maintain” or “we mix”.

- In the Conclusion part, conclusions of the work should be strongly stated rather than a general overview of the study.

 

- In the manuscript, consecutive references should be separated with a comma (i.e. [1, 2]). For example, in line 34 or in line 53.

Moderate editing of English language required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf